Parallel Deep Hypergraph Generation

Implementation of the tools described in the thesis work in computer science "Modelli Generativi per Ipergrafi mediante l'Utilizzo di Tecniche di Deep Learning".

Content

• How to Use
  • Generative Workflow
  • Training Pipe
  • Hypergraph Analysis Tool
• Model Architecture

How to Use

Generation Workflow

The hypergraph generation process can be performed by running the following command from the console. The parameters accepted by the command are:

• dataset: the dataset to use, every file in the data folder is an available option for this command.
• processes: amount of processes to use.
• n: the quantity of hyperedges to generate.

python3 generate.py --dataset "email-Eu" -n 25027 --processes 32

The generated hypergraph can be found in the generated folder.

Training Pipe

python3 train.py --dataset "email-Eu" -epochs 250 --device "cuda"

Hypergraph Analysis Tool

the project includes an analysis tool for the generated hypergraphs. In particular it is useful for producing results relating to the following structural patterns of hypergraph decompositions: Giant Connected Component, Clustering Coefficient, Effective Diameter, Number of Closed Triangles, Singular Value Distribution, Degree Distribution.

python3 analyze.py --dataset "email-Eu" -k 1    # Node-level analysis
python3 analyze.py --dataset "email-Eu" -k 2    # Edge-level
python3 analyze.py --dataset "email-Eu" -k 3    # Triangle-level
python3 analyze.py --dataset "email-Eu" -k 4    # 4clique-level

Model Architecture